BACKGROUND: To develop and assess the technical validity of new computer-aided diagnostic software (CAD) for automated analyses of optical coherence tomography (OCT) images for the purpose of screening for neovascular age-related macular degeneration. METHODS: Artificial visual techniques were used to develop the CAD in two steps: normalisation and feature vector extraction from OCT images; and training and classification by means of decision trees. Technical validation was performed by a retrospective study design based on OCT images randomly extracted from clinical charts. Images were classified as normal or abnormal to serve for screening purposes. Sensitivity, specificity, positive predictive values and negative predictive values were obtained. RESULTS: The CAD was able to quantify image information by working in the perceptually uniform hue-saturation-value colour space. Particle swarm optimisation with Haar-like features is suitable to reveal structural features in normal and abnormal OCT images. Decision trees were useful to characterise normal and abnormal images using feature vectors obtained from descriptive statistics of detected structures. The sensitivity of the CAD was 96% and the specificity 92%. CONCLUSIONS: This new CAD for automated analysis of OCT images offers adequate sensitivity and specificity to distinguish normal OCT images from those showing potential neovascular age-related macular degeneration. These results will enable its clinical validation and a subsequent cost-effectiveness assessment to be made before recommendations are made for population-screening purposes.
BACKGROUND: To develop and assess the technical validity of new computer-aided diagnostic software (CAD) for automated analyses of optical coherence tomography (OCT) images for the purpose of screening for neovascular age-related macular degeneration. METHODS: Artificial visual techniques were used to develop the CAD in two steps: normalisation and feature vector extraction from OCT images; and training and classification by means of decision trees. Technical validation was performed by a retrospective study design based on OCT images randomly extracted from clinical charts. Images were classified as normal or abnormal to serve for screening purposes. Sensitivity, specificity, positive predictive values and negative predictive values were obtained. RESULTS: The CAD was able to quantify image information by working in the perceptually uniform hue-saturation-value colour space. Particle swarm optimisation with Haar-like features is suitable to reveal structural features in normal and abnormal OCT images. Decision trees were useful to characterise normal and abnormal images using feature vectors obtained from descriptive statistics of detected structures. The sensitivity of the CAD was 96% and the specificity 92%. CONCLUSIONS: This new CAD for automated analysis of OCT images offers adequate sensitivity and specificity to distinguish normal OCT images from those showing potential neovascular age-related macular degeneration. These results will enable its clinical validation and a subsequent cost-effectiveness assessment to be made before recommendations are made for population-screening purposes.
Authors: Maximilian W M Wintergerst; Thomas Schultz; Johannes Birtel; Alexander K Schuster; Norbert Pfeiffer; Steffen Schmitz-Valckenberg; Frank G Holz; Robert P Finger Journal: Transl Vis Sci Technol Date: 2017-07-18 Impact factor: 3.283
Authors: Alex M Santos; Anselmo C Paiva; Adriana P M Santos; Steve A T Mpinda; Daniel L Gomes; Aristófanes C Silva; Geraldo Braz; João Dallyson S de Almeida; Marcelo Gattas Journal: Biomed Eng Online Date: 2018-10-23 Impact factor: 2.819